THE PROCESS IN MOTION

What does heat make inside the bean?

Start the scene or choose a stage. The animation simplifies the mechanism; what follows gives the details and the available references.

▶ Play the process01Maillard reaction02Sugar alone caramelizes03Strecker, and what degrades04What remains at the end

Maillard reaction

Free amino acids and reducing sugars react together: this is the Maillard reaction, described by Louis-Camille Maillard in the early 20th century. It yields nitrogenous heterocycles, aromas and melanoidins, the brown pigments that make the color: yellow, then tawny, then light brown. Scott Rao calls it self-sustaining above 160 °C, an order of magnitude rather than a constant, since Maillard is a family of hundreds of reactions.

Read the stages without animation

1. Maillard reactionFree amino acids and reducing sugars react together: this is the Maillard reaction, described by Louis-Camille Maillard in the early 20th century. It yields nitrogenous heterocycles, aromas and melanoidins, the brown pigments that make the color: yellow, then tawny, then light brown. Scott Rao calls it self-sustaining above 160 °C, an order of magnitude rather than a constant, since Maillard is a family of hundreds of reactions.
2. Sugar alone caramelizesSugar alone melts and darkens, from light brown to dark brown, around 171 °C according to Rao. Sucrose, the main fuel, is consumed: it is not found in the cup. Its fragmentation yields organic acids, acetic, formic, lactic, glycolic, which make the bright acidity of light roasts before disappearing. Caramelizing does not sweeten: the sweet taste decreases, bitterness increases.
3. Strecker, and what degradesIn the wake of Maillard, Strecker degradation turns amino acids into aldehydes and pyrazines. Chlorogenic acids degrade with the degree: quinic acid, astringent and bitter, lactones, then guaiacol and 4-vinylguaiacol, hence the smoky and spicy notes. Trigonelline breaks down into pyridines and nicotinic acid.
4. What remains at the endPyrazines peak around 250 °C then are incorporated into melanoidins; 2-furfurylthiol, the odor of roasted coffee par excellence, binds to them too, which contributes to staling. Caffeine is thermostable: its proportion rises as the bean gets lighter. Acrylamide, formed early, peaks in light roasts then declines, because it degrades faster than it forms.

To remember: Heat creates nothing from nothing: it consumes the sugars and amino acids of the green to make color, aromas and bitterness.

In history: Liebig, then Thurber, believed that very dark coffees no longer had caffeine; it is thermostable, and the belief is false.

Read the full article and its sources ↓

The two families of reactions are those of the kitchen. The first is the Maillard reaction, described by Louis-Camille Maillard at the beginning of the 20th century. Free amino acids and reducing sugars react together to give nitrogenous heterocycles, aromas and brown melanoidins. These are what make the color: yellow, then tawny, then light brown.

The second is caramelization, sugar alone melting and darkening, from light brown to dark brown. In the wake of Maillard, Strecker degradation turns amino acids into aldehydes and pyrazines.

Sucrose is the main fuel, and it is consumed: it is not found again in the cup. Its fragmentation also gives organic acids, acetic, formic, lactic, glycolic, which explain the bright acidity of light roasts and its disappearance afterward.

Chlorogenic acids, for their part, degrade with the degree, especially at the “Italian” level. They give quinic acid (astringency, bitterness), lactones and, via caffeic and ferulic acids, guaiacol and 4-vinylguaiacol, hence the smoky and spicy notes. Trigonelline, for its part, breaks down into pyridines and nicotinic acid.

Two molecules are worth following to the end. The pyrazines (roasted, earthy) peak at about 250 °C, then are incorporated into the melanoidins. 2-Furfurylthiol, the “roasted coffee” par excellence, is born from sugars and sulfur compounds, then binds to the melanoidins too, which contributes to staling.

Caffeine, for its part, is thermostable. Liebig's belief, taken up by Thurber (“in the over-dark ones there is no caffeine”) and by Walsh, is therefore false: the losses are limited to a slight sublimation.

Vosloo 2017, § 2.2.1.1, p. 16-17; Sciencedirect, Effect of roasting on chemical composition of coffee, 2025; Wu et al., Food Science & Nutrition, 2022; Hofmann et al., J. Agric. Food Chem., 2001 (thiols and melanoidins); Thurber-Walsh digest, which is false; LC ch. 16b

Two undesirable compounds form in the same reactions. Acrylamide, from asparagine and sugars, appears early, peaks in light roasts and then decreases because it degrades faster than it forms; Regulation (EU) 2017/2158, applicable since April 11, 2018, sets “benchmark levels” (not limits) of 400 µg/kg for roasted coffee and 850 µg/kg for instant, requires determining the critical roasting conditions and building their control into a prerequisite program of good manufacturing practices, exempts the roaster who sells at retail or only to the local retail trade, recalls that canephora contains more than arabica and suggests enzymatic treatment with asparaginase.

Furan and the methylfurans, from sugars, amino acids and fatty acids, make coffee the leading contributor to adult exposure along with cereals; EFSA concluded on October 25, 2017 that there was a health concern (liver) without being able to set a threshold, by margin of exposure. The level of evidence is that of an agency and a regulation: solid on the levels, open on the real risk at consumption doses.

Regulation (EU) 2017/2158, annexes I and IV (EUR-Lex, accessed 09/07/2026); EFSA, press release of 10/25/2017; Sciencedirect, Insights into acrylamide and furanic compounds in coffee, 2024

The thresholds, the sugar that disappears, and the caffeine that concentrates

The section describes the reactions well and gives no onset threshold, that is, no temperature from which they begin to count. Rao gives two, with the sequence that links them, and he adds on sugar a remark that goes against everyone's intuition.

Two thresholds, and the caution that goes with them

Maillard reactions become self-sustaining above 320 °F, that is 160 °C; caramelization begins at about 340 °F, that is 171 °C. Eleven degrees separate the two, which explains why they are always seen together even though they have neither the same reactants nor the same products. These are probe readings, like all those on this plate: Rao reasons on a machine where first crack falls at about 375 °F, 191 °C, according to his answer to a reader on Home-Barista, when most drums read it at about 200 °C; on those, the same thresholds would probably read some ten degrees higher. Online trade guides, moreover, have the Maillard phase begin at yellowing, at about 150 to 160 °C probe temperature.

The wording of the first threshold deserves to be reread with caution, and the plate must say so if it takes it up. Maillard is not a reaction, it is a family of hundreds of reactions between sugars and amino acids. To write that such a family becomes self-sustaining beyond a single threshold is a strong wording for a phenomenon that has none, and the footnote calls that Rao attaches to it refer to references absent from the excerpt consulted. The figure is useful as an order of magnitude, not as a physical constant.

The mid-roast runaway, finally explained

Between his two thresholds, Rao inserts a causal chain that the plate lacks and that is worth more than the two numbers. Once the internal temperature is high enough for most of the moisture to be gone, the temperature of the bean rises faster, because evaporation no longer takes energy from it: turning water from liquid into vapor costs a great deal of heat, and as long as that lasts, the bean is held back.

As soon as this brake disappears, the rise accelerates, which accelerates Maillard, which produces still more heat. This is one of the reasons why aromatic development runs away in the middle of the roast, and it is also why this part of the curve is the hardest to hold without the rate of rise climbing back up.

Rao offers in passing the most telling culinary comparison in the corpus: to understand what Maillard contributes, it is enough to compare a roast meat and a boiled meat. The second has cooked, it has not browned, and it lacks exactly what roasting brings to coffee.

Sugar: a disagreement, and a surprise

The plate says that sucrose is the main fuel and that its fragmentation yields organic acids. Rao adds an element that runs against intuition: caramelization, which is a pyrolysis, that is, a decomposition by heat alone without any other reagent, lowers the sweet taste and raises bitterness. According to him this is the main reason why light roasts are sweeter and dark ones more bitter.

Caramelizing sugar, then, is not making it sweeter, it is destroying it while making something else, and the word caramel misleads everyone here, including the cupping sheets.

Münchow, for his part, draws a curve with only two axes, acidity going down and bitterness going up as the roast advances. Sweetness does not appear in it. This is not a head-on contradiction, it is an absence, and it matters: one makes sweetness a dimension that decreases with the degree, the other does not represent it at all. Neither measures anything, and yet this would be one of the few easily measurable things here, since the residual sugars of a roasted coffee can be assayed.

Caffeine, and the arithmetic the plate forgets

The plate says that caffeine is thermostable, that is, that it does not degrade at roasting temperatures, and stops there. Rao follows the reasoning to the end. Since the quantity of caffeine is practically unchanged and the bean loses mass, the proportion of caffeine relative to weight increases during the roast.

The consequence is counterintuitive and it is arithmetic, not chemical: if all one's coffees are prepared with the same ratio of water mass to ground coffee mass, dark roasts give a more caffeinated beverage than light ones. The precision about mass is decisive and it explains why the opposite belief survives: measured with a spoon, thus by volume, a dark roast puts less matter in the filter, since the bean has swollen and become lighter. Two dosing methods, two opposite results, and the same chemistry behind them.

Scott Rao, The Coffee Roaster's Companion, 2014, p. 18 for the two thresholds, the causal chain, the comparison of roast and boiled meat and the caramelization that reduces the sweet taste, p. 19 for caffeine and the consequence on dosing by mass; Morten Münchow, Roasting Foundation, CoffeeMind, figure 5, p. 15, for the two-axis curve of acidity and bitterness; Home-Barista, thread “What is Baked Coffee?”, accessed 10/21/2026, for the first crack at 375 °F assumed in Rao's book; Barista Hustle, How to Roast Coffee, and Green Plantation, accessed 10/21/2026, for yellowing at about 150 °C.